1. |
- Fu, Keren, et al.
(author)
-
Deepside: A general deep framework for salient object detection
- 2019
-
In: Neurocomputing. - : Elsevier BV. - 0925-2312 .- 1872-8286. ; 356, s. 69-82
-
Journal article (peer-reviewed)abstract
- Deep learning-based salient object detection techniques have shown impressive results compared to con- ventional saliency detection by handcrafted features. Integrating hierarchical features of Convolutional Neural Networks (CNN) to achieve fine-grained saliency detection is a current trend, and various deep architectures are proposed by researchers, including “skip-layer” architecture, “top-down” architecture, “short-connection” architecture and so on. While these architectures have achieved progressive improve- ment on detection accuracy, it is still unclear about the underlying distinctions and connections between these schemes. In this paper, we review and draw underlying connections between these architectures, and show that they actually could be unified into a general framework, which simply just has side struc- tures with different depths. Based on the idea of designing deeper side structures for better detection accuracy, we propose a unified framework called Deepside that can be deeply supervised to incorporate hierarchical CNN features. Additionally, to fuse multiple side outputs from the network, we propose a novel fusion technique based on segmentation-based pooling, which severs as a built-in component in the CNN architecture and guarantees more accurate boundary details of detected salient objects. The effectiveness of the proposed Deepside scheme against state-of-the-art models is validated on 8 benchmark datasets.
|
|
2. |
- Boulund, Fredrik, et al.
(author)
-
Computational and Statistical Considerations in the Analysis of Metagenomic Data
- 2018
-
In: Metagenomics: Perspectives, Methods, and Applications. - 9780081022689 ; , s. 81-102
-
Book chapter (other academic/artistic)abstract
- In shotgun metagenomics, microbial communities are studied by random DNA fragments sequenced directly from environmental and clinical samples. The resulting data is massive, potentially consisting of billions of sequence reads describing millions of microbial genes. The data interpretation is therefore nontrivial and dependent on dedicated computational and statistical methods. In this chapter we discuss the many challenges associated with the analysis of shotgun metagenomic data. First, we address computational issues related to the quantification of genes in metagenomes. We describe algorithms for efficient sequence comparisons, recommended practices for setting up data workflows and modern high-performance computer resources that can be used to perform the analysis. Next, we outline the statistical aspects, including removal of systematic errors and how to identify differences between microbial communities from different experimental conditions. We conclude by underlining the increasing importance of efficient and reliable computational and statistical solutions in the analysis of large metagenomic datasets.
|
|
3. |
- Wiqvist, Samuel, et al.
(author)
-
Partially Exchangeable Networks and architectures for learning summary statistics in Approximate Bayesian Computation
- 2019
-
In: Proceedings of the 36th International Conference on Machine Learning. - : PMLR. ; 2019-June, s. 11795-11804
-
Conference paper (peer-reviewed)abstract
- We present a novel family of deep neural architectures, named partially exchangeable networks (PENs) that leverage probabilistic symmetries. By design, PENs are invariant to block-switch transformations, which characterize the partial exchangeability properties of conditionally Markovian processes. Moreover, we show that any block-switch invariant function has a PEN-like representation. The DeepSets architecture is a special case of PEN and we can therefore also target fully exchangeable data. We employ PENs to learn summary statistics in approximate Bayesian computation (ABC). When comparing PENs to previous deep learning methods for learning summary statistics, our results are highly competitive, both considering time series and static models. Indeed, PENs provide more reliable posterior samples even when using less training data.
|
|
4. |
- Abarenkov, Kessy, et al.
(author)
-
Protax-fungi: A web-based tool for probabilistic taxonomic placement of fungal internal transcribed spacer sequences
- 2018
-
In: New Phytologist. - : Wiley. - 0028-646X .- 1469-8137. ; 220:2, s. 517-525
-
Journal article (peer-reviewed)abstract
- © 2018 New Phytologist Trust. Incompleteness of reference sequence databases and unresolved taxonomic relationships complicates taxonomic placement of fungal sequences. We developed Protax-fungi, a general tool for taxonomic placement of fungal internal transcribed spacer (ITS) sequences, and implemented it into the PlutoF platform of the UNITE database for molecular identification of fungi. With empirical data on root- and wood-associated fungi, Protax-fungi reliably identified (with at least 90% identification probability) the majority of sequences to the order level but only around one-fifth of them to the species level, reflecting the current limited coverage of the databases. Protax-fungi outperformed the Sintax and Rdb classifiers in terms of increased accuracy and decreased calibration error when applied to data on mock communities representing species groups with poor sequence database coverage. We applied Protax-fungi to examine the internal consistencies of the Index Fungorum and UNITE databases. This revealed inconsistencies in the taxonomy database as well as mislabelling and sequence quality problems in the reference database. The according improvements were implemented in both databases. Protax-fungi provides a robust tool for performing statistically reliable identifications of fungi in spite of the incompleteness of extant reference sequence databases and unresolved taxonomic relationships.
|
|
5. |
-
- 2019
-
Journal article (peer-reviewed)
|
|
6. |
- Martinsson, John, et al.
(author)
-
Automatic blood glucose prediction with confidence using recurrent neural networks
- 2018
-
In: CEUR Workshop Proceedings. - : CEUR. ; 2148, s. 64-68
-
Conference paper (peer-reviewed)abstract
- Low-cost sensors continuously measuring blood glucose levels in intervals of a few minutes and mobile platforms combined with machine-learning (ML) solutions enable personalized precision health and disease management. ML solutions must be adapted to different sensor technologies, analysis tasks and individuals. This raises the issue of scale for creating such adapted ML solutions. We present an approach for predicting blood glucose levels for diabetics up to one hour into the future. The approach is based on recurrent neural networks trained in an end-to-end fashion, requiring nothing but the glucose level history for the patient. The model outputs the prediction along with an estimate of its certainty, helping users to interpret the predicted levels. The approach needs no feature engineering or data pre-processing, and is computationally inexpensive.
|
|
7. |
- Mashad Nemati, Hassan, 1982-, et al.
(author)
-
Bayesian Network Representation of Meaningful Patterns in Electricity Distribution Grids
- 2016
-
In: 2016 IEEE International Energy Conference (ENERGYCON). - : IEEE. - 9781467384636
-
Conference paper (peer-reviewed)abstract
- The diversity of components in electricity distribution grids makes it impossible, or at least very expensive, to deploy monitoring and fault diagnostics to every individual element. Therefore, power distribution companies are looking for cheap and reliable approaches that can help them to estimate the condition of their assets and to predict the when and where the faults may occur. In this paper we propose a simplified representation of failure patterns within historical faults database, which facilitates visualization of association rules using Bayesian Networks. Our approach is based on exploring the failure history and detecting correlations between different features available in those records. We show that a small subset of the most interesting rules is enough to obtain a good and sufficiently accurate approximation of the original dataset. A Bayesian Network created from those rules can serve as an easy to understand visualization of the most relevant failure patterns. In addition, by varying the threshold values of support and confidence that we consider interesting, we are able to control the tradeoff between accuracy of the model and its complexity in an intuitive way. © 2016 IEEE
|
|
8. |
- Medved, Dennis
(author)
-
Applications of Machine Learning on Natural Language Processing and Biomedical Data
- 2017
-
Licentiate thesis (other academic/artistic)abstract
- Machine learning is ubiquitous in today’s society, with promising applicationsin the field of natural language processing (NLP), so that computers can handlehuman language better, and within the medical community, with the promiseof better treatments. Machine learning can be seen as a subfield of artificialintelligence (AI), where AI is used to describe a machine that mimics cognitivefunctions that humans associate with other human minds, such as learning orproblem solving.In this thesis we explore how machine learning can be used to improve classification of picture, by using associated text. We then shift our focus to biomedical data, specifically heart transplantation patients. We show how the data can be represented as a graph database using the resource description framework (RDF).After that we use the data with logistic regression and the Spark framework, toperform feature search to predict the survival probability of the patients. In thetwo last articles we use artificial neural networks (ANN) first to predict patientsurvival, and compare it with a logistic regression approach, and last to predict the outcome of patients awaiting heart transplantation.We plan to do simulation of different allocation policies, for donor hearts, usingthese kind of ANNs, to be able to asses their impact on predicted earned survivaltime.
|
|
9. |
- Mogren, Olof, 1980, et al.
(author)
-
Character-based Recurrent Neural Networks for Morphological Relational Reasoning
- 2017
-
In: Proceedings of the First Workshop on Subword and Character Level Models in NLP. - Stroudsburg, PA, United States : Association for Computational Linguistics.
-
Conference paper (peer-reviewed)abstract
- We present a model for predicting word forms based on morphological relational reasoning with analogies. While previous work has explored tasks such as morphological inflection and reinflection, these models rely on an explicit enumeration of morphological features, which may not be available in all cases. To address the task of predicting a word form given a demo relation (a pair of word forms) and a query word, we devise a character-based recurrent neural network architecture using three separate encoders and a decoder. We also investigate a multiclass learning setup, where the prediction of the relation type label is used as an auxiliary task. Our results show that the exact form can be predicted for English with an accuracy of 94.7%. For Swedish, which has a more complex morphology with more inflectional patterns for nouns and verbs, the accuracy is 89.3%. We also show that using the auxiliary task of learning the relation type speeds up convergence and improves the prediction accuracy for the word generation task.
|
|
10. |
- Norinder, Ulf, 1956-, et al.
(author)
-
Conformal prediction to define applicability domain : A case study on predicting ER and AR binding
- 2016
-
In: SAR and QSAR in environmental research (Print). - : Taylor & Francis. - 1062-936X .- 1029-046X. ; 27:4, s. 303-316
-
Journal article (peer-reviewed)abstract
- A fundamental element when deriving a robust and predictive in silico model is not only the statistical quality of the model in question but, equally important, the estimate of its predictive boundaries. This work presents a new method, conformal prediction, for applicability domain estimation in the field of endocrine disruptors. The method is applied to binders and non-binders related to the oestrogen and androgen receptors. Ensembles of decision trees are used as statistical method and three different sets (dragon, rdkit and signature fingerprints) are investigated as chemical descriptors. The conformal prediction method results in valid models where there is an excellent balance in quality between the internally validated training set and the corresponding external test set, both in terms of validity and with respect to sensitivity and specificity. With this method the level of confidence can be readily altered by the user and the consequences thereof immediately inspected. Furthermore, the predictive boundaries for the derived models are rigorously defined by using the conformal prediction framework, thus no ambiguity exists as to the level of similarity needed for new compounds to be in or out of the predictive boundaries of the derived models where reliable predictions can be expected.
|
|